You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

172 lines
6.5 KiB

from typing import Dict, List, Type
from django.db.models import QuerySet
from . import writers
from .catalog import CSV, JSON, JSONL, FastText
from .comments import Comments
from .formatters import (
DictFormatter,
FastTextCategoryFormatter,
Formatter,
JoinedCategoryFormatter,
ListedCategoryFormatter,
RenameFormatter,
TupledSpanFormatter,
)
from .labels import BoundingBoxes, Categories, Labels, Relations, Segments, Spans, Texts
from data_export.models import DATA, ExportedExample
from projects.models import Project, ProjectType
def create_writer(file_format: str) -> writers.Writer:
mapping = {
CSV.name: writers.CsvWriter(),
JSON.name: writers.JsonWriter(),
JSONL.name: writers.JsonlWriter(),
FastText.name: writers.FastTextWriter(),
}
if file_format not in mapping:
ValueError(f"Invalid format: {file_format}")
return mapping[file_format]
def create_formatter(project: Project, file_format: str) -> List[Formatter]:
use_relation = getattr(project, "use_relation", False)
# text tasks
mapper_text_classification = {DATA: "text", Categories.column: "label"}
mapper_sequence_labeling = {DATA: "text", Spans.column: "label"}
mapper_seq2seq = {DATA: "text", Texts.column: "label"}
mapper_intent_detection = {DATA: "text", Categories.column: "cats"}
mapper_relation_extraction = {DATA: "text"}
# image tasks
mapper_image_classification = {DATA: "filename", Categories.column: "label"}
mapper_bounding_box = {DATA: "filename", BoundingBoxes.column: "bbox"}
mapper_segmentation = {DATA: "filename", BoundingBoxes.column: "segmentation"}
mapper_image_captioning = {DATA: "filename", Texts.column: "label"}
# audio tasks
mapper_speech2text = {DATA: "filename", Texts.column: "label"}
mapping: Dict[str, Dict[str, List[Formatter]]] = {
ProjectType.DOCUMENT_CLASSIFICATION: {
CSV.name: [
JoinedCategoryFormatter(Categories.column),
JoinedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_text_classification),
],
JSON.name: [
ListedCategoryFormatter(Categories.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_text_classification),
],
JSONL.name: [
ListedCategoryFormatter(Categories.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_text_classification),
],
FastText.name: [FastTextCategoryFormatter(Categories.column)],
},
ProjectType.SEQUENCE_LABELING: {
JSONL.name: [
DictFormatter(Spans.column),
DictFormatter(Relations.column),
DictFormatter(Comments.column),
RenameFormatter(**mapper_relation_extraction),
]
if use_relation
else [
TupledSpanFormatter(Spans.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_sequence_labeling),
]
},
ProjectType.SEQ2SEQ: {
CSV.name: [
JoinedCategoryFormatter(Texts.column),
JoinedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_seq2seq),
],
JSON.name: [
ListedCategoryFormatter(Texts.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_seq2seq),
],
JSONL.name: [
ListedCategoryFormatter(Texts.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_seq2seq),
],
},
ProjectType.IMAGE_CLASSIFICATION: {
JSONL.name: [
ListedCategoryFormatter(Categories.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_image_classification),
],
},
ProjectType.SPEECH2TEXT: {
JSONL.name: [
ListedCategoryFormatter(Texts.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_speech2text),
],
},
ProjectType.INTENT_DETECTION_AND_SLOT_FILLING: {
JSONL.name: [
ListedCategoryFormatter(Categories.column),
TupledSpanFormatter(Spans.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_intent_detection),
]
},
ProjectType.BOUNDING_BOX: {
JSONL.name: [
DictFormatter(BoundingBoxes.column),
DictFormatter(Comments.column),
RenameFormatter(**mapper_bounding_box),
]
},
ProjectType.SEGMENTATION: {
JSONL.name: [
DictFormatter(Segments.column),
DictFormatter(Comments.column),
RenameFormatter(**mapper_segmentation),
]
},
ProjectType.IMAGE_CAPTIONING: {
JSONL.name: [
ListedCategoryFormatter(Texts.column),
ListedCategoryFormatter(Comments.column),
RenameFormatter(**mapper_image_captioning),
]
},
}
return mapping[project.project_type][file_format]
def select_label_collection(project: Project) -> List[Type[Labels]]:
use_relation = getattr(project, "use_relation", False)
mapping: Dict[str, List[Type[Labels]]] = {
ProjectType.DOCUMENT_CLASSIFICATION: [Categories],
ProjectType.SEQUENCE_LABELING: [Spans, Relations] if use_relation else [Spans],
ProjectType.SEQ2SEQ: [Texts],
ProjectType.IMAGE_CLASSIFICATION: [Categories],
ProjectType.SPEECH2TEXT: [Texts],
ProjectType.INTENT_DETECTION_AND_SLOT_FILLING: [Categories, Spans],
ProjectType.BOUNDING_BOX: [BoundingBoxes],
ProjectType.SEGMENTATION: [Segments],
ProjectType.IMAGE_CAPTIONING: [Texts],
}
return mapping[project.project_type]
def create_labels(project: Project, examples: QuerySet[ExportedExample], user=None) -> List[Labels]:
label_collections = select_label_collection(project)
labels = [label_collection(examples=examples, user=user) for label_collection in label_collections]
return labels
def create_comment(examples: QuerySet[ExportedExample], user=None) -> List[Comments]:
return [Comments(examples=examples, user=user)]